AI Agent Integration

AI coding agents (Claude Code, Cursor, Copilot) often need access to secrets — API keys, database credentials, tokens — without those values leaking into prompts, logs, or generated code.

Coldrune provides two integration paths:

MCP (recommended)

The Coldrune MCP server exposes 15 tools over the Model Context Protocol. Agents can read secrets, manage projects, and query audit logs through structured tool calls.

Benefits:

• Secrets stay server-side — agents receive values only when explicitly requested
• Agents can create projects, environments, and manage infrastructure
• Full audit trail of agent actions
• Works with any MCP-compatible client

CLI

Agents can also call the coldrune CLI directly. Authenticate with a service account API key:

export COLDRUNE_API_KEY=cr_sa_...
coldrune secret get DB_PASSWORD --org my-org --project api --env dev

Or use coldrune run to inject secrets into a subprocess:

coldrune run --org my-org --project api --env dev -- ./migrate.sh

Authentication for agents

Method	Use case
Service account (COLDRUNE_API_KEY)	CI/CD, automated scripts, agents
Session token	Interactive human use

Service accounts are recommended for agents. They:

• Don’t require interactive login
• Can be scoped to a specific project
• Always require explicit ACL rules (deny-by-default)
• Leave clear audit trails identifying the service account

Create one:

coldrune service-account create --org my-org --name claude-agent
coldrune acl grant --org my-org --subject claude-agent --project "*" --env "*" --permission read

Machine-readable docs

Coldrune serves machine-readable documentation at:

• /llms.txt — curated index of key documentation pages
• /llms-full.txt — full documentation concatenated as plain text

See llms.txt & AGENTS.md for details.